FedShard: Federated Unlearning with Efficiency Fairness and Performance Fairness

The introduction of FedShard marks a pivotal advancement in federated learning, particularly in addressing the rights of clients to be forgotten. As federated unlearning becomes increasingly important, FedShard stands out as the first algorithm designed to ensure both efficiency fairness and performance fairness. By effectively removing the data contributions of clients who leave, it accelerates the unlearning process significantly—1.3 to 6.2 times faster than retraining from scratch and 4.9 times faster than existing exact unlearning methods. The theoretical analysis and numerical evaluations confirm its effectiveness, demonstrating that FedShard not only enhances unlearning efficiency but also mitigates risks associated with unfairness, such as cascaded leaving and poisoning attacks. This dual focus on fairness and performance is crucial as it addresses the challenges faced by decentralized clients, ensuring that the unlearning process is equitable and efficient.